Article Details

Abstract

To address urban traffic congestion, this paper proposes an intelligent traffic signal control method based on Enhanced Dueling DQN. This method integrates the Dueling network architecture with enhanced techniques such as prioritized experience replay, learning rate scheduling, and gradient clipping, constructing an efficient deep reinforcement learning model to achieve global optimization of traffic states. Based on the SUMO traffic simulation platform, experimental results demonstrate that, compared to the basic DQN and its variants, the Enhanced Dueling DQN algorithm achieves slightly lower average speed while outperforming in metrics such as average reward, average queue length, and average waiting time. The Enhanced Dueling DQN algorithm significantly improves traffic flow efficiency, providing an effective technical solution for intelligent traffic signal control with high academic reference value and great engineering application potential.

Keywords

prioritized experience replay; intelligent transportation; learning rate scheduling; gradient clipping; Enhanced Dueling DQN

References

1. [1] Zhao Fenglong, Zhong Bo, Li Yilong Research on Intelligent Traffic Signal Control Algorithm Based on Deep Reinforcement Learning [J]. Journal of Jiujiang University (Natural Science Edition), 2025, 40 (4): 13-17, 97
2. [2] Zhang Xiaohan, Zhang Junle, Yan Fei Traffic signal control method based on historical information iterative learning control [J]. Computer Engineering and Design, 2024, 45 (9): 2649-2657
3. [3] Ren Anni, Zhou Dake, Feng Jinhao, etc Deep reinforcement learning based on attention mechanism for traffic signal control [J]. Computer Application Research, 2023, 40 (2): 430-434
4. [4] Zhang Chen, Jiang Wenying, Chen Siyuan, etc Multi agent path planning based on double-layer DQN [J]. Chinese Journal of Image and Graphics, 2023, 28 (7): 2167-2181
5. [5] Xie Mengwei, Chen Guang, Wang Yang, etc Traffic signal control based on decision and execution separation using DQN algorithm [J]. Journal of Zhejiang University of Science and Technology, 2025, 37 (5): 526-536
6. [6] Ye Baolin, Chen Dong, Liu Chunyuan, etc Traffic signal control method based on DuelingDoubleDQN [J]. Computer Measurement and Control, 2024, 32 (7): 154-161
7. [7] Wang Feng, Wang Sai Single intersection signal optimization method based on DuelingDQN+PER [J]. Journal of People's Public Security University of China (Natural Science Edition), 2025, 31 (2): 50-58
8. [8] Wang Dingsheng, Ding Lei Intelligent Traffic Signal Control Algorithm Based on MP-DDQN [J]. Journal of Shaanxi University of Science and Technology, 2025, 43 (2): 196-202214
9. [9] Wei Min, Cai Changjian Traffic signal control based on improved deep reinforcement learning [J]. Computer Engineering and Design, 2025, 46 (3): 927-933
10. [10] Zhang Wei, Li Yujun, Xie Wenwen, etc Generative SAC algorithm based on priority experience replay and its application [J]. Journal of Jilin University (Science Edition), 2025, 63 (6): 1713-1722
11. [11] Zhao Hongxing, Wang Yujie, Nie Jianglong, etc Deep reinforcement learning signal continuous control of intersections based on cellular deduction and multi-step decision-making [J]. Journal of Transportation Engineering, 2025, 25 (4): 296-310
12. [12] Li Qian, Zhou Wencai, Li Yongqi Adaptive Signal Control of Single Intersection Based on Improved GBML-DQN Algorithm [J]. Journal of Beijing Jiaotong University, 2025, 49 (6): 110-125

Funding

Joint Fund Project of Jingzhou Science and Technology Bureau and College of Arts and Sciences, Yangtze University (2023LHX02)

About the authors

Zhu Rongtao (1982-), male, Han ethnicity, from Jingzhou, Hubei, associate professor, master's degree, research direction: artificial intelligence and embedded systems; He Zhaoxia (1984-), female, Han ethnicity, from Huanggang, Hubei, associate professor, master's degree, research direction: signal and information processing.